Plain Review: B2B AI Support Stack for Slack, Ari & Sidekick

Plain bundles inquiries from Slack, Teams, Discord, and email into a single inbox, and pairs a customer-facing AI agent called Ari with an agent-facing copilot called Sidekick to handle first response through investigation. It's a B2B customer support platform, priced by seats plus AI credits starting at $35/month.

This article is based on information published on plain.com's official site, pricing page, product pages, changelog, official blog, and GraphQL documentation as of September 23, 2026. It covers how Plain works, pricing, the developer API, and where it fits for small Japanese teams.

Plain's tagline is "The AI support stack built for B2B." Where traditional help desks are built around the ticket, Plain is built around meeting customers where they already are. Inquiries from Slack Connect shared channels, Slack Enterprise Grid, Slack communities, Microsoft Teams, Discord, email, a support portal, a contact form, live chat, and a help center all land in a single Plain inbox, organized as threads, Customer Cards, and workflows. Plain says it brings "hundreds of Slack channels" into one inbox with SLA tracking and breach alerts built for Slack, and describes the connection itself as three steps — connect your workspace, add the app to a channel, and messages sync automatically — taking about 30 seconds. From there, Ari handles routine questions first, with cited replies, and hands off to a human agent when it can't resolve something. Human agents then work with Sidekick to pull in context from Linear, Jira, CRM tools, monitoring tools, and custom MCP servers while drafting a reply.

Ari is the AI agent that talks to customers and resolves issues. It reads the full thread and drafts cited replies grounded in help center articles, internal docs, and the knowledge base. Plain states that "every answer links directly to its source and can be reviewed before it reaches the customer."

Before going live, teams can test Ari in two ways: Playground, a private chat for asking questions and watching Ari's reasoning, and Shadow Mode, which observes Ari's behavior against live traffic without actually sending messages. Tone and escalation rules are configurable, and Ari hands off the moment a human agent joins the conversation. Teams can also route only specific customer segments to Ari, based on revenue, plan, or any custom data.

An August 18, 2026 release added Playground, Shadow Mode, citation tracking, and custom instructions. A September 21, 2026 update added the ability for Ari to ask clarifying questions when a customer message is vague, plus configurable handoff behavior for when a customer tags a team member. Ari is included starting from the entry-level Foundation plan. Whether Ari can run on a company's own LLM isn't stated on the official pages (third-party articles mention this, but it's unconfirmed).

Sidekick launched on July 13, 2026, per the official blog post "Meet Sidekick." Co-founder Simon Rohrbach frames the problem it addresses this way: "Customer questions are rarely answerable from the conversation alone" — the time sink is the research before you can even write a reply, not the writing itself. A typical resolution flow looks like this: Ari handles a routine question first, with cited replies, and hands off to a human agent if it can't resolve the issue or the customer tags a teammate. Sidekick then gathers context across the conversation, engineering systems, documentation, and connected tools, reasons across those sources, and drafts a conclusion plus suggested actions. Finally, a human agent reviews that draft and hits Send.

Reusable Custom Skills shipped on July 27, 2026, and Sidekick Home — a hub for session management, a list of connected tools, and starting investigations — shipped on July 12, 2026. A September 21, 2026 update added the ability to build a workflow (triggers, conditions, actions) from a natural-language description. Plain's own examples of what Sidekick can do include summarizing overnight queue changes, identifying the Linear issue tied to a request, tracking a customer's setup progress during onboarding, investigating the code behind an error, drafting help center articles, analyzing customer sentiment and account health, and producing daily briefings.

One constraint is explicit in the official docs: "Sidekick can suggest a reply, but a team member needs to review it and hit Send. It can't message a customer on its own, and that behavior can't currently be changed." Sidekick is on by default across every paid plan.

AI usage is metered with credits, and each paid plan gets a monthly allotment that resets at the start of the billing cycle; extra purchased credits roll over to the following month. According to a Featurebase article, credits are metered by "effort, not volume" — a short summary costs little, while a deep investigation across multiple tools costs more. There's no published conversion rate, and once credits run out, Sidekick reportedly pauses until the next reset (per that article). Also per Featurebase, seat caps are said to be 5 seats max on Foundation and 10 on Horizon, though the official pricing page doesn't state any cap.

A startups program offers Horizon at 50% off — a flat $6,000 for 12 months and 10 seats. Eligibility requires VC funding of $5M or less raised, a support team of 10 people or fewer, and not already being a Plain customer. The process is: book a demo, get an eligibility check, then a 2-week free trial.

On pricing philosophy, Plain states that "because AI usage is credit-based and tied to the plan, cost follows the plan rather than scaling with a per-resolution charge." Plain's own comparison content also claims that "roughly half of Intercom Fin users cite per-resolution AI pricing as a pain point" (Plain's claim, not independently verified here).

Worth noting: n8n's result comes from building its own AI agent on top of Plain's API, not from using Ari or Sidekick out of the box. Funding raised, customer count, and whether Japanese UI or Japanese-language support exist aren't stated on the official pages — the admin interface and documentation are English-only.

Plain is API-first, with a GraphQL endpoint at https://core-api.uk.plain.com/graphql/v1 (POST). Authentication uses an Authorization: Bearer <API_KEY> header with permission scopes such as customer:read. The schema is published at https://core-api.uk.plain.com/graphql/v1/schema.graphql, with customers, threads, tenants, tiers, and labels as the main objects. Plain's own minimal example looks like this.

Here's that same query run as a curl example (the official docs don't publish a curl example directly, so this adapts the minimal query above into a runnable request).

Through the API's "machine user" concept — where an AI agent acts as a user — Plain also connects to third-party AI agents such as Decimal, Duckie AI, Inkeep, and Parahelp. Plain's own comparison content describes "BYOA on Frontier: connect your own model and tools via API and MCP," meaning the Frontier plan supports connecting your own model and tools over the API and MCP. A September 22, 2026 changelog entry lets a machine user reply as a human team member across Slack, email, and the timeline. Webhook details, rate limits, and SDK names aren't covered on the pages referenced here — check the rest of the documentation for those.

This comparison reflects claims from Plain's own comparison content and isn't a definitive verdict either way — check Zendesk, Intercom, Front, Kustomer, and others directly for current pricing and features. Building a support AI in-house with a local LLM and RAG is also a viable alternative.

- A good fit for dev shops and SaaS teams already talking to customers over Slack Connect — worth weighing against a lighter setup like routing inquiries to Slack, LINE, and Chatwork notifications
- The admin interface and documentation are in English; whether a Japanese UI exists isn't stated officially
- Ari's response quality in Japanese is unverified — test it in Shadow Mode before trusting it in production
- AI credit conversion rates aren't published, so the practical move is to start on Foundation and measure actual burn
- Third-party reporting of seat caps (5 on Foundation, 10 on Horizon) is worth checking against your team size before committing to a plan
- For phone- or email-heavy support, a traditional help desk or a simple form-plus-Slack-notification setup may be cheaper. Teams with more engineers may value the issue-tracker integration described in What Is Linear?; for how to split work between humans and AI, see Using AI for customer support, and for building a support AI in-house, see Building a RAG customer support AI with Ollama and OpenClaw

Plain is a B2B support platform that meets customers where they already are — Slack, Teams, Discord, email — and pairs Ari's first-line responses with Sidekick's research support for human agents. Pricing runs on seats plus AI credits starting at $35/month, with a GraphQL API and BYOA for developer-led extension.

That said, a lot remains unstated officially: Japanese-language support, whether Ari can run on a company's own LLM, and how AI credits actually convert to usage. Before committing, it's worth testing real behavior and credit burn with Shadow Mode and the Foundation plan's free trial.

- https://www.plain.com/
- https://www.plain.com/pricing
- https://www.plain.com/ari
- https://www.plain.com/sidekick
- https://www.plain.com/blog/meet-sidekick
- https://www.plain.com/changelog
- https://www.plain.com/customers
- https://www.plain.com/startups
- https://www.plain.com/docs/graphql/introduction
- Third-party: https://www.featurebase.app/blog/plain-pricing